Hi, thanks for your great paper and source code!
I am sorry that I am confused about the dataset_dict in FewX/fewx/data/dataset_mapper.py.
I found that the value of dataset_dict['annotations'] only have one category-id( it may be some objects but also belong to one category-id).
In my view, I thought the dataloader is related to the json file FewX/datasets/coco/new_annotations/final_split_non_voc_instances_train2017.json, however when I observed the FewX/datasets/coco/1_split_filter.py, I thought the file just removes novel classes's annotations. Is it right?
But I am still confused that why the dataset_dict['annotations'] can not have more category_id? It is different from the common setting of detectron2 dataset_mapper after I debug it.
Could anyone give me some help?
Hi, thanks for your great paper and source code! I am sorry that I am confused about the
dataset_dict
inFewX/fewx/data/dataset_mapper.py
. I found that the value ofdataset_dict['annotations']
only have one category-id( it may be some objects but also belong to one category-id). In my view, I thought thedataloader
is related to the json fileFewX/datasets/coco/new_annotations/final_split_non_voc_instances_train2017.json
, however when I observed theFewX/datasets/coco/1_split_filter.py
, I thought the file just removes novel classes's annotations. Is it right?But I am still confused that why the
dataset_dict['annotations']
can not have more category_id? It is different from the common setting ofdetectron2
dataset_mapper after I debug it. Could anyone give me some help?